智能系统算法与模型库
收藏资源简介：
本数据集面向动态环境下多无人系统的感知、协同决策与数据驱动控制研究，覆盖无人车集群编队、无人机自主探索与形式化规划，以及生物数据驱动建模等多个典型应用方向。数据集由多个相互关联的子任务数据构成，包括多无人车动态环境目标识别与集群编队感知数据、多无人机随机环境下的线性时序逻辑（LTL）约束规划与信息安全控制数据，以及面向基因调控网络构建的多模态生物测序数据。各子数据在任务目标与数据结构上相互独立，但在数据组织、标注规范与质量控制流程上保持统一标准，以提升数据复用性与跨任务研究的可比性。在空间范围上，数据集同时覆盖真实物理环境、半实物仿真环境与纯仿真场景；在时间维度上记录多段连续任务全过程，精度达到视频帧级或状态转移级。感知类数据主要以RGB视频、多视角同步图像、三维位姿信息及多源传感器记录形式存储，支持二维像素级目标标注与基于标定算法计算的三维空间位姿。规划与控制类数据基于概率模型与形式化任务描述生成，完整记录状态转移、控制输入、策略执行轨迹及算法对比结果，支持任务可满足性与系统不透明性分析。生物数据部分则包含统一预处理的真实测序数据与标准流程生成的仿真数据，覆盖细胞类型反卷积、单细胞数据填充与基因调控网络推断等关键任务。数据采集与生成过程中综合采用人工标注、算法自动标定及策略对比实验等方式，并通过多视角同步采集、标定校验与一致性检查等质量控制措施保证数据可靠性与标注一致性。该数据集可广泛用于动态目标识别与跟踪、多智能体协同感知、无人机自主探索、形式化规划与安全控制，以及数据驱动建模与生物网络推断等研究方向，具有较高的科研价值与工程应用潜力。
This dataset is targeted at research on perception, collaborative decision-making and data-driven control of multi-unmanned systems in dynamic environments, covering multiple typical application directions including unmanned ground vehicle (UGV) cluster formation, unmanned aerial vehicle (UAV) autonomous exploration and formal planning, and biological data-driven modeling. This dataset consists of multiple interrelated sub-task datasets, including perception data for UGV dynamic environment target recognition and cluster formation, data for UAV linear temporal logic (LTL) constrained planning and information security control in stochastic environments, and multimodal biological sequencing data for gene regulatory network (GRN) construction. Each sub-dataset is independent in terms of task objectives and data structure, but adheres to unified standards for data organization, annotation specifications and quality control processes, to improve data reusability and comparability for cross-task research. In terms of spatial scope, the dataset covers real physical environments, hardware-in-the-loop (HITL) simulation environments and pure simulation scenarios; in terms of temporal dimension, it records multiple complete continuous task execution processes with precision down to video frame level or state transition level. Perception-related data is mainly stored in the form of RGB videos, multi-view synchronized images, 3D pose information and multi-source sensor records, supporting 2D pixel-level target annotation and 3D spatial poses calculated using calibration algorithms. Planning and control-related data is generated based on probabilistic models and formal task descriptions, fully recording state transitions, control inputs, policy execution trajectories and algorithm comparison results, supporting task satisfiability and system opacity analysis. The biological data section includes uniformly preprocessed real sequencing data and simulation data generated via standard procedures, covering key tasks such as cell type deconvolution, single-cell data imputation and gene regulatory network inference. Comprehensive methods including manual annotation, automatic algorithm calibration and policy comparison experiments are employed during data collection and generation, and quality control measures such as multi-view synchronous collection, calibration verification and consistency inspection are implemented to ensure data reliability and annotation consistency. This dataset can be widely used in research directions such as dynamic target recognition and tracking, multi-agent collaborative perception, UAV autonomous exploration, formal planning and security control, as well as data-driven modeling and biological network inference, with high scientific research value and engineering application potential.
